48f67d26ad
License: MIT Signed-off-by: Jakub Sztandera <kubuxu@protocol.ai>
165 lines
3.4 KiB
Go
165 lines
3.4 KiB
Go
package peermgr
|
|
|
|
import (
|
|
"context"
|
|
"sync"
|
|
"time"
|
|
|
|
"github.com/filecoin-project/lotus/node/modules/dtypes"
|
|
"go.uber.org/fx"
|
|
|
|
host "github.com/libp2p/go-libp2p-core/host"
|
|
net "github.com/libp2p/go-libp2p-core/network"
|
|
peer "github.com/libp2p/go-libp2p-core/peer"
|
|
dht "github.com/libp2p/go-libp2p-kad-dht"
|
|
|
|
logging "github.com/ipfs/go-log"
|
|
)
|
|
|
|
var log = logging.Logger("peermgr")
|
|
|
|
const (
|
|
MaxFilPeers = 32
|
|
MinFilPeers = 12
|
|
)
|
|
|
|
type MaybePeerMgr struct {
|
|
fx.In
|
|
|
|
Mgr *PeerMgr `optional:"true"`
|
|
}
|
|
|
|
type PeerMgr struct {
|
|
bootstrappers []peer.AddrInfo
|
|
|
|
// peerLeads is a set of peers we hear about through the network
|
|
// and who may be good peers to connect to for expanding our peer set
|
|
//peerLeads map[peer.ID]time.Time // TODO: unused
|
|
|
|
peersLk sync.Mutex
|
|
peers map[peer.ID]time.Duration
|
|
|
|
maxFilPeers int
|
|
minFilPeers int
|
|
|
|
expanding bool
|
|
|
|
h host.Host
|
|
dht *dht.IpfsDHT
|
|
|
|
notifee *net.NotifyBundle
|
|
}
|
|
|
|
func NewPeerMgr(h host.Host, dht *dht.IpfsDHT, bootstrap dtypes.BootstrapPeers) *PeerMgr {
|
|
pm := &PeerMgr{
|
|
h: h,
|
|
dht: dht,
|
|
bootstrappers: bootstrap,
|
|
|
|
peers: make(map[peer.ID]time.Duration),
|
|
|
|
maxFilPeers: MaxFilPeers,
|
|
minFilPeers: MinFilPeers,
|
|
}
|
|
|
|
pm.notifee = &net.NotifyBundle{
|
|
DisconnectedF: func(_ net.Network, c net.Conn) {
|
|
pm.Disconnect(c.RemotePeer())
|
|
},
|
|
}
|
|
|
|
h.Network().Notify(pm.notifee)
|
|
|
|
return pm
|
|
}
|
|
|
|
func (pmgr *PeerMgr) AddFilecoinPeer(p peer.ID) {
|
|
pmgr.peersLk.Lock()
|
|
defer pmgr.peersLk.Unlock()
|
|
pmgr.peers[p] = time.Duration(0)
|
|
}
|
|
|
|
func (pmgr *PeerMgr) GetPeerLatency(p peer.ID) (time.Duration, bool) {
|
|
pmgr.peersLk.Lock()
|
|
defer pmgr.peersLk.Unlock()
|
|
dur, ok := pmgr.peers[p]
|
|
return dur, ok
|
|
}
|
|
|
|
func (pmgr *PeerMgr) SetPeerLatency(p peer.ID, latency time.Duration) {
|
|
pmgr.peersLk.Lock()
|
|
defer pmgr.peersLk.Unlock()
|
|
if _, ok := pmgr.peers[p]; ok {
|
|
pmgr.peers[p] = latency
|
|
}
|
|
|
|
}
|
|
|
|
func (pmgr *PeerMgr) Disconnect(p peer.ID) {
|
|
if pmgr.h.Network().Connectedness(p) == net.NotConnected {
|
|
pmgr.peersLk.Lock()
|
|
defer pmgr.peersLk.Unlock()
|
|
delete(pmgr.peers, p)
|
|
}
|
|
}
|
|
|
|
func (pmgr *PeerMgr) Run(ctx context.Context) {
|
|
tick := time.NewTicker(time.Second * 5)
|
|
for {
|
|
select {
|
|
case <-tick.C:
|
|
pcount := pmgr.getPeerCount()
|
|
if pcount < pmgr.minFilPeers {
|
|
pmgr.expandPeers()
|
|
} else if pcount > pmgr.maxFilPeers {
|
|
log.Debug("peer count about threshold: %d > %d", pcount, pmgr.maxFilPeers)
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
func (pmgr *PeerMgr) getPeerCount() int {
|
|
pmgr.peersLk.Lock()
|
|
defer pmgr.peersLk.Unlock()
|
|
return len(pmgr.peers)
|
|
}
|
|
|
|
func (pmgr *PeerMgr) expandPeers() {
|
|
if pmgr.expanding {
|
|
return
|
|
}
|
|
pmgr.expanding = true
|
|
go func() {
|
|
defer func() {
|
|
pmgr.expanding = false
|
|
}()
|
|
ctx, cancel := context.WithTimeout(context.TODO(), time.Second*30)
|
|
defer cancel()
|
|
|
|
pmgr.doExpand(ctx)
|
|
}()
|
|
}
|
|
|
|
func (pmgr *PeerMgr) doExpand(ctx context.Context) {
|
|
pcount := pmgr.getPeerCount()
|
|
if pcount == 0 {
|
|
if len(pmgr.bootstrappers) == 0 {
|
|
log.Warn("no peers connected, and no bootstrappers configured")
|
|
return
|
|
}
|
|
|
|
log.Info("connecting to bootstrap peers")
|
|
for _, bsp := range pmgr.bootstrappers {
|
|
if err := pmgr.h.Connect(ctx, bsp); err != nil {
|
|
log.Warnf("failed to connect to bootstrap peer: %s", err)
|
|
}
|
|
}
|
|
return
|
|
}
|
|
|
|
// if we already have some peers and need more, the dht is really good at connecting to most peers. Use that for now until something better comes along.
|
|
if err := pmgr.dht.Bootstrap(ctx); err != nil {
|
|
log.Warnf("dht bootstrapping failed: %s", err)
|
|
}
|
|
}
|